Wind Energy Forecasting
Overview:
A business problem to find the best place to set up a wind farm in the drive towards cleaner and greener energy production was given as a course project. Traditional statistical models such as AR, MA, ARMA, ARIMA and SARIMA were tried along with some ML algorithms(NaiveForecaster, KNN Regressor, Ensemble methods, etc.). ACF(autocorrelation) and PACF(partial autocorrelation) plots were used to decide the lag orders for the AR and MA components and in turn decide the model best suited for cause. Finally, Statistical Quality Control analysis was concluded for wind energy generation potential for the four states in the dataset. The results from statistical models, machine learning models and SQC analysis were then collated to recommend a state for setting up the wind farm.
Personal Note:
This was my first foray into time series forecasting and my first time applying traditional statistical approaches for solving a problem. Upon conclusion of the project I had found a new source of interest in time series forecasting and I started reading up more about research problems in this domain. My next goal is to be able to leverage GNNs to solve some multi-variate time series problems.
Technical Details
- Language: Python
- Libraries: sklearn, sktime, statsmodels